NASA Technical Reports Server (NTRS) 20170001233: Planetary Probe Entry Atmosphere Estimation Using Synthetic Air Data System Planetary Probe Entry Atmosphere Estimation Using Synthetic Air Data System

1.1 MB·English

by  NASA Technical Reports Server (NTRS)

Planetary Probe Entry Atmosphere Estimation Using
Synthetic Air Data System
Christopher D. Karlgaard∗
Analytical Mechanics Associates, Inc., Hampton, VA
Mark Schoenenberger†
NASA Langley Research Center, Hampton, VA
This paper develops an atmospheric state estimator based on inertial acceleration and angular
rate measurements combined with an assumed vehicle aerodynamic model. The approach utilizes
the full navigation state of the vehicle (position, velocity, and attitude) to recast the vehicle aero-
dynamic model to be a function solely of the atmospheric state (density, pressure, and winds).
Force and moment measurements are based on vehicle sensed accelerations and angular rates.
These measurements are combined with an aerodynamic model and a Kalman-Schmidt filter to
estimate the atmospheric conditions. The new method is applied to data from the Mars Science
Laboratory mission, which landed the Curiosity rover on the surface of Mars in August 2012. The
results of the new estimation algorithm are compared with results from a Flush Air Data Sensing
algorithm based on onboard pressure measurements on the vehicle forebody. The comparison in-
dicates that the new proposed estimation method provides estimates consistent with the air data
measurements, without the use of pressure measurements. Implications for future missions such
as the Mars 2020 entry capsule are described.
Nomenclature
C = state-parameter covariance
F = aerodynamic force, N
F = linearization of f with respect to x
f = process model
G = linearization of f with respect to u
g = gravitational acceleration, m/s2
H = linearization of h with respect to x
h = measurement model
I = identity matrix
I = vehicle inertia, kg-m2
J = linearization of h with respect to u
K = filter gain
k = integer time index
M = aerodynamic moment, N-m
M = Mach number
P = covariance of x
p = static pressure, Pa
s
Q = process noise spectral density
∗SupervisingEngineer,SeniorMemberAIAA.
†AerospaceEngineer,AtmosphericFlightandEntrySystemsBranch,SeniorMemberAIAA.
1of16
AmericanInstituteofAeronauticsandAstronautics
Q˜ = process noise covariance
R = measurement error covariance matrix
(cid:60) = specific gas constant, J/kg-K
r = radius, m
T = atmospheric temperature, K
v , v , v = vehicle planet-relative north, east, and down velocity components, m/s
n e d
w , w , w = north, east, and down wind velocity components, m/s
n e d
x = atmospheric state vector
Θ = longitude, rad
θ, φ, ψ = vehicle pitch, roll, and yaw attitude angles, rad
Λ = declination, rad
λ = aerodatabase uncertainty factors
µ = gravitational parameter, m3/s2
ν = vehicle inertial state
ρ = density, kg/m3
Φ = state transition matrix
I. Introduction
NASA has developed an Entry, Descent, and Landing (EDL) technology development roadmap1 to guide in-
vestment strategies for increased EDL capabilities and robustness. One area of emphasis is on the development of
precisionlandingcapabilitiesachievedthroughimprovedenvironment/atmospherecharacterizationandEDLinstru-
mentation for validation of engineering models and ground testing procedures. One approach that can be used to
address these areas is the implementation of a Flush Air Data Sensing (FADS) system, which utilizes an array of
pressure ports installed in the vehicle forebody to measure the pressure distribution during entry. These pressure
measurementscanbeprocessedtoestimatethefreestreamaerodynamicstate(suchasflowangles,Machnumber,and
dynamic pressure), atmospheric conditions (density, pressure, and winds) and vehicle aerodynamics. These sensors
can be used for post-flight trajectory reconstruction and model validation, but also have the potential to be used
to augment the on-board flight control system by providing estimates of density and winds if the data processing
algorithms can be implemented in real-time.
Incorporation of pressure measurements into the heatshield of an entry vehicle is not trivial. Implementation
of a reliable FADS system typically involves hardware development and qualification, optimization of pressure port
layout,sensorcalibration,groundtestingtoensurepressureportandthermalprotectionsystemintegrity,etc.,which
can be costly. Additionally, a FADS system is subject to risks of sensor failures and other hardware anomalies.
Typically,planetaryentryatmospherereconstructionisperformedusingvehicleaerodynamicmodelswhenFADS
instrumentation is not available. The concept of utilizing vehicle aerodynamic models for entry probe atmosphere
estimation is not new. In fact, the concept goes back to a 1963 NASA report4 in which a proposal was made
to “invert” the entry physics problem by solving for atmospheric density, given acceleration measurements and a
model of vehicle drag. The concept was further developed to estimate aerodynamic flow angles in later papers,
such as Ref. 5. The methodology was validated using entry vehicle test flights on Earth, with known atmospheric
conditions,6 and was subsequently applied to the reconstruction of the atmosphere of Mars based on measurements
fromtheVikingentryprobes.7 SinceViking,theapproachhasbeenutilizedonvirtuallyeveryplanetaryatmospheric
entry reconstruction to date. A recent thesis8 documents the development and history of the algorithm, from its
inception in the early 1960s to the most recent application of the method for the Mars Science Laboratory mission.9
The previous approaches make use of measured accelerations and known aerodynamic models to solve for the
freestream conditions by first computing density from the axial force coefficient and axial acceleration. The recon-
structed density is then used to integrate the hydrostatic equation to estimate static pressure. The ratio of normal
to axial and side to axial forces are then utilized to estimate the angles of attack and sideslip, respectively. The
process can be iterated at each instant in time to improve the estimates; for instance, a combined inner and outer
loop is implemented in Ref. 9 for solving the aerodynamic flow angles (inner loop) and freestream atmospheric con-
ditions (outer loop). Note that the algorithm is completely deterministic in nature, although uncertainties in the
reconstructed quantities can be computed using linear covariance analysis techniques.8,9 Estimation of winds from
aerodynamics has not been addressed directly; some approaches instead rely on a post-processing method such as
that proposed in Ref. 10 for computing winds from flow angles and the navigation state.
2of16
AmericanInstituteofAeronauticsandAstronautics
A recently proposed alternative to a traditional FADS system is a technique known as a Synthetic Air Data
Sensing (SADS) system. The SADS approach makes use of an aerodynamic model of the vehicle, combined with ac-
celerationandangularratemeasurementsonanonboardInertialMeasurementUnit(IMU)toestimatethefreestream
aerodynamicandatmosphericconditions. ASADSsystemcanbeusedeitherasabackup,orasalow-costalternative
to a traditional FADS system.
This paper extends the SADS concept to the planetary probe entry vehicle atmosphere estimation problem. The
method developed in this paper is most similar to that of Refs. 12 and 13, in which a model of the vehicle dynamics
is combined with an inertial navigation system to produce estimates of the atmospheric winds using a Kalman
filter method for improved navigation and flight control. One important drawback of these approaches is that the
atmosphericdensityandpressurearenotestimatedinthefilter;insteadtheseareassumedtobeknownasafunction
ofaltitude. Thisassumptionisnotsuitableforplanetaryprobeentryestimationinwhichtheatmosphericproperties
are highly uncertain. In this paper, the concept of atmospheric estimation aided by a vehicle aerodynamics model
is extended to estimate freestream density and pressure in addition to winds. This new approach is developed for
planetary entry state estimation, but it is anticipated that the method can be applied to other applications such as
high speed aircraft.
II. Atmospheric State Estimation
The new approach for SADS-based state estimation proposed in this paper makes use of the full state from the
on-board navigation system (position, velocity, and attitude) combined with the vehicle aerodynamic database to
directly estimate the atmospheric conditions. The estimation algorithm is aided by both high-fidelity atmospheric
models tabulated vs. altitude and simplified atmospheric models that are propagated along the trajectory within
the algorithm.
A. Measurement Model
Thisworkassumesthatanaerodynamicmodeloftheentrycapsuleisavailable,inwhichforceandmomentcoefficients
can be calculated from a given flight condition (angle of attack, sideslip, Mach number, etc.). It is assumed the
aerodynamicmodelproducesoutputsinsomeknowncoordinateframe, suchasthatshowninFigure1. Dimensional
aerodynamic forces and moments acting on the vehicle are then computed using the relations
Figure 1: Body Coordinate System
3of16
AmericanInstituteofAeronauticsandAstronautics
 
1  −CA(α,β,M,λ) 
F = ma= ρV2S C (α,β,M,λ) (1)
2 Y
 −C (α,β,M,λ) 
Z
 
1  Cl(α,β,M,λ) 
M = Iω˙ +ω×Iω = ρV2Sb C (α,β,M,λ) (2)
2 m
 C (α,β,M,λ) 
n
whereλisavectorcomprisedoftheaerodatabaseuncertaintyfactors,whichareacollectionofaddersandmultipliers
used to perturb the aerodynamics.
Given that dynamic pressure is a function of vehicle velocity, winds, and density; Mach number is a function of
velocity, winds, density, and pressure; aerodynamic flow angles are functions of velocity, winds, and attitude, the
aerodynamic force and moment model can be recast into the form
ma = F(x,u) (3)
Iω˙ +ω×Iω = M(x,u) (4)
(cid:124) (cid:124)
where x = [ρ,p ,w ,w ,w ] is the atmospheric state, u = [ν,λ] are model parameters, and the vehicle planet-
s n e d
(cid:124)
relative navigation state is ν =[r,Θ,Λ,v ,v ,v ,φ,θ,ψ] .
n e d
High-fidelity atmosphere models can also be incorporated into the state estimate as prior information. These
atmosphere models are assumed to be computationally intensive such that implementation within the algorithm is
infeasible. Instead, the high-fidelity atmosphere model data can be incorporated using table look-ups where the
atmospheric conditions and uncertainties are tabulated as a function of altitude along some nominal trajectory. The
model of this form produces an estimate of the atmospheric conditions, along with an associated error covariance
matrix.
Theaerodynamicforceandmomentandhigh-fidelityatmospheremodelscanbecombinedintoasingleexpression,
(cid:124)
z =h(x,u),whereh=[F(x,u),M(x,u),x] . NotethatthemeasurementerrorcovariancematrixRisafunction
of both sensor and mass property uncertainties.
B. Process Model
Alow-fidelitymodelofthechangeinatmosphericconditionsalongthetrajectorycanbederivedfrombasicidealized
relations such as the hydrostatic equation and the perfect gas law. Such simplified relationships are suitable for
implementationinthealgorithmforpropagatingtheatmosphericstate estimateforwardbetweenaerodynamicmea-
surements, which are assumed to occur at a reasonably high rate (several samples per second) along the trajectory.
Since the simplified model involves idealized approximations, uncertainties in the model can be accounted for with
process noise.
A model for the rate of change in static pressure can be found by rewriting the hydrostatic equation as the time
derivative of pressure along a given trajectory, namely
p˙ =ρgv (5)
s d
Similarly, a model for the rate of change in density along the trajectory can be derived from the perfect gas law,
withtheassumptionthattheatmosphereislocallyisothermal(T˙ ≈0)betweenmeasurementsamples. Theequation
is of the form
p˙ p˙ ρ gv ρ2
ρ˙ = s = s = d (6)
(cid:60)T p p
s s
A reasonable simplified model for the rate of change in atmospheric winds is to assume a random walk model
where the deterministic portion of the model is simply w˙ = w˙ = w˙ = 0. Thus, the low fidelity model can be
n e d
written in the form
x˙ =f(x,u)+η (7)
where η is a process uncertainty term that is assumed to be zero mean with spectral density Q, and
 
gv ρ2/p
 ρdgvd s 
f(x,u)= 0 (8)
 0 
0
4of16
AmericanInstituteofAeronauticsandAstronautics
The continuous model in Eq. (7) can be transformed to a discrete model of the form,
x =x +f(x ,u )∆t (9)
k+1 k k k
which is suitable for propagation between measurements.
C. Data Fusion Algorithm
The atmospheric state estimate can be determined from a fusion of the available data sources, including the aero-
dynamic force and moment measurements, and information from the low and high fidelity models. The proposed
algorithm is in the form of an Iterated Extended Kalman-Schmidt Filter (IEKSF)15 incorporating the process and
measurementmodelsdescribedabove. TheIEKSFapproachincorporatesparameteruncertaintiesintheprocessand
measurement models, thus producing a realistic state covariance estimate.
The algorithm is structured as a predictor-corrector method, in which state estimates are propagated between
measurement samples using the relations
x¯ = xˆ +f(x¯ ,u¯ )∆t (10)
k+1 k k k
P¯ = Φ Pˆ Φ(cid:124)+Φ Cˆ G(cid:124)+G C(cid:124)Φ(cid:124)+G Ω G(cid:124)+Q˜ (11)
k+1 k k k k k k k k k k k k k
C¯ = Φ Cˆ +G Ω (12)
k+1 k k k k
where Φ is the state transition matrix and Q˜ is the discrete-time process noise covariance. These quantities can
k k
be jointly calculated from the Van Loan matrix integral,16 given by
(cid:32)(cid:34) (cid:35) (cid:33) (cid:34) (cid:35) (cid:34) (cid:35)
−F Q X X X Φ−1Q˜
exp k k ∆t = 11 12 = 11 k k (13)
(cid:124)
0 F 0 X 0 Φ
k 22 k
whichleadstotheresultΦ =X(cid:124) andQ˜ =Φ X . Assumingareasonablyfastintegrationstep,thesequantities
k 22 k k 12
can be approximated by Φ ≈I+F ∆t and Q˜ ≈Q ∆t.
k k k k
The measurement update is an iterative process to solve a nonlinear least squares regression problem, involving
the measurement and the state prediction as observations. The process is given by the equations
S = H P¯ H(cid:124) +H C¯ L(cid:124)+L C¯(cid:124)H(cid:124) +R (14)
k,i k,i k,i k,i k,i k k k k k,i k
(cid:104) (cid:105)
K = P¯ H(cid:124) +C¯ L(cid:124) S−1 (15)
k,i k k,i k k k,i
xˆ = x¯ +K [z −h(xˆ ,u )−H (x¯ −xˆ )] (16)
k,i+1 k k,i k k,i k k,i k k,i
Equations(14)–(16)areiterateduntilconvergenceoruntilreachingaprescribediterationlimit. Aftertheiteration
is complete, the state covariance and state-parameter covariance matrices can be computed as
Pˆ = P¯ −K S K(cid:124) (17)
k k k k k
Cˆ = C¯ −K (cid:2)H C¯ +L Ω (cid:3) (18)
k k k k k k k
The Kalman-Schmidt filter approach has the advantage of providing optimal state estimates that account for
systematic parameter uncertainties to produce a realistic state covariance estimate.
D. Aerodynamic State Transformations
The atmospheric state (winds, pressure, and density) are outputs of the proposed SADS data processing algorithm.
The atmospheric state can readily be combined with the INS state solution to produce estimates of aerodynamic
states, including angle of attack, sideslip, Mach number, and dynamic pressure. Uncertainties can be mapped from
theatmosphericandINSstatesintotheaerodynamicstatesthroughlinearcovarianceanalysis. Theequationsofthe
transformation from atmospheric and INS states to aerodynamic states are readily available in various sources such
as Ref. 18 and are not repeated here.
III. Application to Mars Science Laboratory
On August 5th 2012, the Mars Science Laboratory (MSL) entry vehicle successfully entered the atmosphere of
Mars and landed the Curiosity rover safely on the surface of the planet in Gale crater. The MSL entry vehicle
5of16
AmericanInstituteofAeronauticsandAstronautics
was comprised of a 70-degree sphere-cone heatshield and backshell consisting of a stack of three truncated cones.
The forebody was similar to the heatshield geometry developed for the Viking Mars landers. Phenolic Impregnated
Carbon Ablator (PICA) was used for the thermal protection system material. The backshell configuration was also
similar to Viking, with a third cone section added to accommodate the parachute volume. The MSL vehicle as-built
outer mold line is shown in Fig. 2(a).19 During most of entry, the capsule used a radial center of mass offset to
fly at an angle of attack (approximately 16 degrees at hypersonic conditions). This attitude produced lift to fly a
guided entry profile, reducing the landing footprint to a much smaller size than any previous Mars mission. To fly
the guided entry, the vehicle carried four pairs of Reaction Control System (RCS) jets to perform maneuvers and
damp rates. The four pairs of jets could be fired rapidly in different combinations to provide control torque about
any axis by modulating the pulses of the jet.
Spacecraft Coordinate Frame
+y . +x
+z
4518.16
1295.39
R12.7
BIP: SC z=0.0 34.43° 119°(543.3)
31°
33.73°
R161.8
53.1°
2874.82 1407.3 22.09°
BS-HS
Sep. Plane
R134.88
924.21 20° R126.75
SR1134.11
(a) AsBuiltMSLOuterMoldLine (b) MEDLI/MEADSGeometry(LookingAft)
Figure 2: Vehicle Geometry
MSL carried with it an instrumentation package designed to measure the aerodynamic and aerothermal environ-
ments during atmospheric entry. This instrumentation package was known as the MSL Entry, Descent, and Landing
Instrumentation (MEDLI),20 which consisted of three major subsystems: the Mars Entry Atmospheric Data System
(MEADS), the MEDLI Integrated Sensor Plugs (MISP), and the Sensor Support Electronics (SSE). The MEADS
consisted of seven pressure transducers connected to flush orifices in the heat shield to measure pressures across
the vehicle forebody. The MISP devices were a system of seven thermocouple and recession sensors that provided
aerothermalmeasurementsoftheheatshieldperformance. TheSSEprovidedpowertothesensors,conditionedtheir
signals, and transmitted the data to storage on the Curiosity rover. The MEDLI sensors provided measurements
that were used for trajectory reconstruction and engineering validation of aerodynamic, atmospheric, and thermal
protectionsystemmodelsinadditiontoEarth-basedsystemstestingprocedures. TheMEDLIdataanditsusagefor
reconstructingtheaerodynamicandaerothermalperformanceoftheMSLentryvehiclearedescribedinRefs.21–29.
The aerodynamic forces and moments were sensed by the on-board IMU in the form of acceleration and angular
rate measurements at a frequency of 200 Hz. The dimensional aerodynamic forces and moments were calculated
from these measurements, combined with mass and inertia models of the vehicle. Forces and moments due to the
RCS firing were subtracted from the total force and moment measurements. The results were found to be sensitive
to vibration and noise associated with RCS firings, so a low-pass optimal Fourier filter with a cut-off frequency of 2
Hz was applied to smooth the data.26 The raw and filtered aerodynamics are shown in Figure 3. Note the increase
in noise around the time of bank reversals, which correlate with the times of RCS thruster commands shown in
Figure 4. The filtered side/normal and pitch/yaw moments are shown in Figure 5.
The SADS estimator was initialized at 580 s in the MSL EDL timeline, corresponding to an altitude of approxi-
mately 66.15 km. The initial atmospheric state was based on pre-flight mesoscale models.32 The air data filter was
implemented to run at a rate of 64 Hz to match the rate of the on-board navigation state. The filter was terminated
at a time of 800 s, at an altitude of 7.10 km. Given the sensitivity of moments to the RCS noise, the filter was set to
ignore moment measurements while the RCS was active and rely solely on force measurements. The estimator made
use of the post-flight reconstructed aerodynamic model described in Ref. 26.
6of16
AmericanInstituteofAeronauticsandAstronautics
(a) AxialForce (b) RollingMoment
(c) SideForce (d) PitchingMoment
(e) NormalForce (f) YawingMoment
Figure 3: Measured Aerodynamic Forces and Moments
7of16
AmericanInstituteofAeronauticsandAstronautics
Figure 4: RCS Firing
(a) SideForce (b) PitchingMoment
(c) NormalForce (d) YawingMoment
Figure 5: Filtered Aerodynamic Forces and Moments
8of16
AmericanInstituteofAeronauticsandAstronautics
Results of the SADS estimator are shown in the following figures. The results are compared to true air data
estimation results from the MEADS pressure measurement data, described in Reference 29. In the following figures,
the pressure transducer-based estimation is labeled FADS, and the new synthetic air data solution developed in this
paper is labeled as SADS.
(a) Density (b) Pressure
Figure 6: Atmosphere
The atmosphericdensityand pressureestimatesare shown inFigure 6. Thetwosolutions give consistent results.
Some differences are apparent in the low supersonic flight regime, below 850 Pa dynamic pressure. These differences
areconsistentwiththosebetweenthepost-flightreconstructedaxialforcecoefficientandthereconciledaerodynamics,
as described in Ref. 26. These differences are most likely attributable to transducer instrumentation errors in low
pressure ranges that are outside their design requirements. Differences in density and pressure estimates are within
0.5% over the range in which the transducers were calibrated (above 850 Pa dynamic pressure).
The estimated winds along the trajectory are shown in Figure 7. The two methods produce consistent estimates,
although the SADS exhibits more noise in the estimate. The noise is likely a combination of higher data rate (64
Hz compared to the 8 Hz FADS solution) and errors due to vibrations in the accelerometer. The mean profile and
general trends follow the FADS solution. This result is important as it indicates the ability of the SADS approach
to mimic a true FADS without the use of pressure sensors. The reconstructed winds from these two methods are
consistent with observed vehicle dynamics and guidance response.34,35
ThedynamicpressureandMachestimatesofareshowninFigure8. Themethodsareinagreementovertheentry
trajectory. Differences in the dynamic pressure and Mach estimates are on the order of 0.5% and 0.05, respectively,
in region where the pressure measurements were calibrated. The aerodynamic flow angle estimates are shown in
Figure 9. The FADS and SADS solutions are in agreement, with differences on the order of 0.25 deg, thus indicating
the SADS method is able to provide an estimate of the vehicle state that is consistent with a true air data system.
9of16
AmericanInstituteofAeronauticsandAstronautics
(a) NorthWind (b) EastWind
(c) DownWind
Figure 7: Winds
(a) DynamicPressure (b) MachNumber
Figure 8: Dynamic Pressure and Mach Number
10of16
AmericanInstituteofAeronauticsandAstronautics